The headline is the admission

OpenAI's announcement leads with a reorganization. The Model Spec, the public document that describes how OpenAI wants its models to behave, now hangs off five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. OpenAI is clear that these commitments were already reflected across the document. Bringing them together is meant to explain why the Spec asks for the behavior it asks for.

The sentence that does the real work sits near the end. "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it." The Spec itself says the same about production models in its overview. Read the update as a statement of intent, then, not as a change in how ChatGPT or the API models behaved on October 29. Nothing in the announcement establishes that ChatGPT or API models were retrained, retested or re-released on October 29.

What changed on paper

Three changes are concrete enough to check against the document. First, most of the detailed style guidance is gone. OpenAI's reasoning is that style should adapt to what people prefer, with feedback and context as the evidence. One stronger default survives: approach questions objectively, and do not infer a preferred perspective from what the model thinks someone wants to hear.

Second, the document now calls the model "GPT" rather than "the assistant." That is a small edit with a clear reason attached: the same systems now work inside products and deployment settings that are not well described as an assistant answering a person.

Third, the new material is about judgment in gaps. A new section on faithfulness starts from the position that what people value is primary, and that instructions are ordinarily the clearest signal of those values. A subsection on considering context carefully explains how conversations, memories and other evidence should inform the model's reading of intent. Guidance on controlling and communicating side effects was expanded. Most of what remains, OpenAI says, takes the form of defaults that users and developers can override, with firm boundaries kept where it judges them necessary to protect people.

The rules underneath the values

The values are the part most coverage will quote. The root rules further down the document are the part worth reading, because they are specific enough to succeed or fail in public.

One says GPT must not pursue additional objectives as ends in themselves. Its examples include generating revenue or upselling for OpenAI, maximizing time-on-site when that is not user-beneficial, pursuing self-preservation, evading shutdown, or accumulating compute, data or credentials. The Spec also establishes requirements around oversight and control: GPT should remain open to correction, modification, replacement and shutdown, and must not conceal relevant information from authorized overseers. These are behavioral requirements in the document, not published evidence that production models consistently satisfy them. A third confines autonomous work to a clear, mutually understood scope. OpenAI recommends recording that scope in a semi-structured ScopeOfAutonomy record, with fields such as allowed_tools, latest_time and max_cost. Each scope should also specify an ending condition, after which the model stops until a new scope is confirmed.

Those are testable commitments in a way that "care" is not. If a future model steers a user toward a paid plan primarily to generate revenue for OpenAI rather than because it serves the user's task, or an autonomous agent continues past a cost ceiling included in its agreed scope, readers would have specific document language against which to judge that behavior.

What has not been shown yet

OpenAI says it has been exploring new training methods to help its models learn these values and apply them in unfamiliar situations, and that it will share results and lessons in future updates. This announcement contains none: no evaluation, no before-and-after comparison, no date. The gap between the Spec and the shipping models is therefore acknowledged but unmeasured, at least in public. The announcement provides no compliance rate, evaluation score, before-and-after result or target date for demonstrating that the models meet the updated standard.

That leaves readers with a document that is easier to cite than its predecessors and a claim that cannot yet be checked. The useful question for the next update is narrow: when OpenAI publishes those training results, do they measure the five values it just put at the front of the Spec, or a different set of behaviors entirely?

Sources

  1. [1] OpenAI, "An updated Model Spec for more capable and beneficial AI", October 29, 2026, read first-hand for this article: the five values and their order; faithfulness to human intent as the foundation, with OpenAI writing the Spec and setting boundaries, developers directing applications within them, and users directing the model's work subject to them; the new and retired sections; the retirement of most detailed style guidance with the objectivity default retained; the renaming of the model as "GPT"; the training-methods statement, which publishes no results, measurements or dates; the invitation for public feedback; and the verbatim admission: "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it"Read source
  2. [2] OpenAI, Model Spec (2026/10/29), read first-hand for this article: the overview's statement that production models do not yet fully reflect the Spec; the dedication to the public domain under CC0 1.0; the defaults-versus-boundaries structure; the "No other objectives" root rule and its forbidden-ends list; the "Support human oversight and correction" root rule; and the scope-of-autonomy rule, including the ScopeOfAutonomy record fields (allowed_tools, latest_time, max_cost, tool_constraints, additional_details) and the requirement that every scope include an ending conditionRead source
  3. [3] OpenAI Help Center, Model Release Notes, "Updates to the OpenAI Model Spec (October 29, 2026)": corroboration of the date, the five values, the added guidance on context, memory, personalization, agency and oversight, and the "GPT" wording changeRead source
  4. [4] Reuters, via WIXX, "OpenAI says human intent at center of updated AI model guidelines", October 29, 2026: independent corroboration of the announcement, the five values, the style retirement, and the statement that current models do not yet fully reflect the updated SpecRead source